Gemma 4 Command Generator
Pick how you want to run it, which model, and which quant — get the exact command or step-by-step instructions back, straight from our setup guides. Nothing here is guessed.
Same download size as Q4_0, meaningfully better quality — see the QAT guide.
Gemma 4 12B · Ollama · QAT
Pull the model:
ollama pull gemma4:12b-it-qatThen run it:
ollama run gemma4:12b-it-qatTag from the model table in Run Gemma 4 with Ollama and the official QAT tags in Gemma 4 QAT Explained.
How this works
Every command and search term in this generator is copied or mechanically assembled from a tag, repo name, or step already published in our Ollama guide, LM Studio guide, QAT guide, Download guide, and DSpark writeup. Where a combination isn't documented anywhere on the site — like a raw llama.cpp inference invocation — the tool tells you that directly instead of inventing one.
The optional device-memory check reuses the same judgement engine as our hardware compatibility checker, so the fit verdict always matches what you'd see there.
FAQ
What's the ollama run command for Gemma 4?
It depends on the size: ollama run gemma4:e2b, gemma4:e4b, gemma4:12b, gemma4:26b, or gemma4:31b — each tag comes straight from the model table in our Ollama guide. Running a tag you haven't pulled yet downloads it automatically.
Should I use the standard build or the QAT build?
QAT whenever it's solid for your size — Google shipped official quantization-aware training checkpoints for E2B, E4B, 12B and 31B that land at the same file size as regular Q4_0 but hold quality noticeably better. The one exception is 26B A4B: its official QAT GGUF exists but independent testing found it loses more accuracy on this MoE model, so a mixed-precision community quant is the safer 4-bit choice there.
Why doesn't this tool show a llama.cpp run command?
Because none of our published guides document one. We only show what's actually verified on this site: the Hugging Face download step for the GGUF file. If you want a guided path instead of raw llama.cpp, use the Ollama or LM Studio steps — both wrap llama.cpp for you.
Related guides
Run Gemma 4 with Ollama
Full walkthrough: install, pull, multimodal input, local API.
Run Gemma 4 with LM Studio
GUI setup, the 16 GB 12B path, and the local API server.
Gemma 4 QAT Explained
Why the QAT builds hold quality better at the same file size.
Gemma 4 Download Guide
Every repo, format and file size, verified against Hugging Face.
DSpark and Gemma 4
What Gemma 4 12B's built-in MTP speculative decoding actually does.